Does Passing Matter?
/By Jared Young
I really want passing to matter. I watch, on average, 968 passes in a soccer game, and I’d like to think that completing them actually means something more than launching the ball into the first row.
Watching a beautiful through ball unhinge a defense is like watching a sun set on the bay. But that beauty doesn’t mean it matters, at least not to the data. Not if you care about winning.
Let me show you the correlation between a team’s pass completion rate and their expected goal difference over the last three years in Major League Soccer:
source: FBREF.com, Americansocceranalysis.com
Not only does it not really matter, but teams with higher pass completion rates actually perform slightly worse. MLS is in another universe compared to the world’s top leagues like the English Premier League, German Bundesliga or La Liga, which have correlations of .71, .89, and .75, respectively.
The big leagues have this relationship due to the difference between the haves and have nots. Pass completion rates are mostly driven by style of play. The most talented teams play a possession oriented brand with shorter, higher percentage passes. The less talented teams have to cede possession, and therefore choose to play more direct, route one soccer from a passing perspective.
Meanwhile, Major League soccer is structured to balance talent across the league as much as possible. That enables teams to experiment with styles, regardless of their skill level. This is something that MLS fans should celebrate, but it sure does make finding the value of passing difficult.
A Deeper Exploration
There are a number of models out there that quantify the difficulty of passes and can help bring context to each pass. ASA’s xPass model assesses the probability that a given pass will be completed. Let’s look at how teams perform relative to the difficulty of the passes they attempt and see if they perform better.
We’re getting somewhere. Controlling for the difficulty of the pass, the better passing teams do have a slight correlation with expected goal difference. But we’re still not close to the big leagues.
But what about that beautiful through ball? Surely that boot helped their team win, right? Perhaps it’s in the individual data.
Again, there are a handful of models that rate a player’s passing ability, but there’s not a lot of public literature on how well that passing ability impacts outcomes. Below is the correlation between an individual’s pass completion rate less their expected pass completion rate and Goals Added (g+). Here we can isolate players that pass better than average and also contribute to goals.
Oh boy. Here we go again. You might think of these first three charts as anti-charts. Looking for a signal, but finding just white space. We can’t find a link between passing competence and the ultimate objective of the game, winning.
The above chart informs us that passers who complete a higher percentage of their passes adjusting for their difficulty have a slightly negative impact on goal creation. I know, I know, this is not a complete assessment. It’s overly simplistic. But shouldn’t an intuitive relationship appear at the highest level in order to be useful?
We could spend hours discussing why this might be the case, preferably over beers: why do our eyes deceive us? Why do 1,000 actions in a match resolve to nothing? And how is it those 1,000 actions simultaneously decide the very outcome itself? What is this thing we’re watching, anyway?
It could also be an example of the limitation of analytics to accurately capture the value of the beauty we see with our own eyes. It could just be that the vast majority of passes are benign, and the constructive moments are lost in the haystack of data. Perhaps we need to identify outlier passes, or better put, the passes made that are outstanding choices compared to what most passers would do in that identical situation. Perhaps there we will find the value. Some projects attempt this, but none question whether better passing algorithms lead to goal creation.
But here we are, and we can’t give up. Passing must mean something in MLS. And the data must be there somewhere.
One of the problems with pass completion rates is that teams tend to play distinct styles that take those rates to the extremes. As addressed above, some teams play very direct soccer, trying to get up the pitch as quickly as possible before the other team can get their defensive shape. Their passes are far riskier than average and their completion rates suffer as a result. The Philadelphia Union, St. Louis City and New York Red Bulls all play this style.
On the other extreme are the teams that want to control the ball. If you don’t have it, you can’t score, is the tiki-taka philosophy. They slowly build their play, working the ball across the pitch waiting for a crack in the defense to exploit. They make shorter, lower risk passes. LAFC is the prime league example.
What if we controlled for this style of play difference? There’s a metric called directness, which sums the progressive distance covered by all passes and divides that by the sum of the total distance covered by all passes (nod to @markrstats for the muse). Let’s see if controlling for directness gives passing metrics any kind of meaning.
source: FBREF.com, @MARKRSTATS
We’ve got a little something here, it appears. When controlling for directness and looking for the most efficient pass completion rates, at a high level we can capture 0.225 goals per game (subtracting the xGD/90 of teams above the efficient frontier line compared to teams below the best fit line). That adds up to a +7.7 goal difference per season. Not bad.
The correlations we’ve been looking at are a bit better too. The correlation of pass completion rate relative to the pass completion rate expected based on the directness of the passes with xGD/90 over the past three seasons is now 0.22. A notable improvement, but hardly a revelation.
The Philadelphia Union season through this new lens
The use of this metric comes to life with a live example. The Philadelphia Union are one of the notably direct teams in the league. Here is a plot of their directness level and pass completion percentage for each game this season.
directness and pass completion percentage for each PHILADELPHIA UNION game this season
This example might be too good, but you can see that when the Union have had an efficient (above average) pass completion percentage relative to their directness, they are undefeated at 7-0-2. For their other seven games, they average one point per game.
Part of the Union’s early season struggles stems from the fact that they were less efficient passing the ball when controlling for their directness. During the Eastern Championship campaign, their directness was 43% while completing 73% of their passes, well above the expected league average of 70%. This season they are still completing 72% of their passes, but only 40% of their passes are direct. Seems like something to watch during the second half of the season.
Let’s spend some time on the lonely dot in the upper left, that Red Bulls game. Talk about an outlier. The Union were over 10% more direct than any other game this season. Two direct teams taking it to the extreme. The Red Bulls directness was an absurd 48.3% as well. It may have been the most direct league game of all time.
It was also terrible to watch. After the match I commented to another Union fan, “That was the worst Union win I’ve ever seen.” But the Union were actually efficient given the ugliness they set out to achieve (or New York forced them into). They were even close to the efficient frontier. Shows what I know.
As lost as I still am, I’m feeling a bit better. The case for passing might be in a dataset somewhere. From here, we have two paths forward. We can probe deeper analytically, perhaps turn this problem over to some new AI, or we can always wait until Major League Soccer is one of the top leagues in the world.
In the meantime, watch a ball game and enjoy the arc of a beautiful pass as it sets over a scrambling center back. Be in awe of the skill it takes to forge that path. There’s some hope it matters.